Forge · run real work

Mission shape agents actually run

SourceA is an AI execution platform powered by Forge — real builds, automations, and agent workflows, with a verifiable receipt on every run. Every job leaves a bounded mission: clear goal, done criteria, and how to verify — not open-ended chat.

How it works under the hood

Output shape (every run)

GOAL: Wire sourcea.app DNS to Cloudflare Pages with clean URLs — no .html in the address bar.

CONTEXT (true now — do not re-derive):
- ALREADY DONE / DEPLOYED (do NOT redo): green-unified dist on disk · _redirects written
- Systems involved: Cloudflare Pages · sourcea.app DNS · publish script

WHAT I WANT: Live site shows extensionless URLs; old .html paths 301 to clean paths.

DONE = curl -I https://sourcea.app/sourcea/proof returns 200 at clean URL; .html URL 301s.

VERIFY: curl -sI https://sourcea.app/sourcea/proof | head · curl -sI https://sourcea.app/sourcea/proof.html | grep -i location

CONSTRAINTS:
- NO rebuild from scratch if dist already correct
- One job only — DNS + publish, not unrelated product work

IF BLOCKED or ambiguous: STOP and tell me what's missing or which decision is mine.

This is the same shape enforced by scripts/prompt_forge_pipeline_v1.py — not a marketing abstraction.